Week 1/24-1/28

＾Making Gnutella-like P2P Systems Scalable￣

　　　The new system proposed, Gia, attempts to solve a basic problem all Gnutella-like systems have: when faced with high aggregate query rate, nodes quickly become overloaded and the system ceases to function satisfactorily. Gia recognizes that most peers have capacity constraints. This issue is addressed in dynamic topology adaptation protocol. This protocol places most nodes within short reach of high capacity nodes, ensuring well-connected (high capacity) nodes actually have the capacity to handle large volume of queries. This design, coupled with One-hop replication, is especially aimed at addressing Gnutella-like systems¨ flooding-based search method, thus protecting peers with low capacity links (modem, isdn etc).
　　　
　　　The author brought up churn rate of large networks when addressing the transient nature of P2P clients. ＾Measured activity in Gnutella and Napster indicates that the median up-time for a node is 60 minutes. For large systems of, say, 100,000 nodes, this implies a churn rate of over 1600 nodes coming and going per minute.￣ Given such a high volume of node entering and leaving the network, I question the usefulness and feasibility of One-hop replication. Over the course of one minute, high capacity nodes must update and flush its neighbor content index 1600 times combined. However, the amount of work at high capacity nodes could be reduced significantly with partial replication at the cost of query efficiency.
　　　
　　　I found the section on DHTS disadvantages to be particularly interesting. It helped to reinforce contents covered in class. The analogy ＾Most queries are for hay, not needles￣ is perfect P2P systems.